You picked Claude Code. Or Cursor. Or whatever agent actually fits how you work, because the agent you use matters more than the editor it lives in. You set it up. You tuned your prompts. You built the muscle memory.
And then you tried to do anything serious with AL.
Yeah.
Here’s the thing about agentic coding that you only really get after doing it for a while: the model is maybe a third of the equation. The tools it has access to are another third. The last third is the feedback loop – how fast and how clearly the agent finds out it just did something stupid.
Take diagnostics away. Take tests away. Now what does your agent have? Vibes. It writes code, decides it looks fine, moves on, and three steps later you are cleaning up the confident hallucinations of a model that very politely told you “good job” based on absolutely nothing.
You would not run a kitchen this way. A chef tastes the sauce. He tastes it again after every adjustment. Take away his tongue and tell him to prepare a dinner for forty – that is roughly what an AL agent feels like when it cannot see a compile error or run a test. The food shows up. It might even look great. But you are not eating it.
So diagnostics and tests are not nice-to-haves. They are the feedback loop. And in AL, that feedback loop was wired into exactly one agent: GitHub Copilot. Until today.
The MCP Bridge for AL Language Model Tools is a new VS Code extension that exposes the AL Language tools as an MCP server. Any MCP-aware agent can now use them. Claude Code. Cursor. Whatever you fire up tomorrow. As long as it speaks MCP, it can build your project, fetch your diagnostics, and run your tests.
That’s it. No new ceremony, no new workflow. You install it, point your agent at it, and the feedback loop closes.
Grab it here: https://marketplace.visualstudio.com/items?itemName=vjeko.vjeko-al-mcp-bridge
Try it with your agent of choice. Tell me what works, what doesn’t, what you want next. Comments are open.
